_Py_atomic_add_*_relaxed

Feature or enhancement

Proposal:

I would like to add relaxed variants of the _Py_atomic_add functions.

The existing add operations are sequentially consistent, which is stronger (and on ARM, measurably more expensive) than necessary for uses like statistics counters and unique ID allocation, where the add must be atomic but does not need to order surrounding memory accesses.

Personally, I'd like to be able to use this in the CinderX JIT, but there are quite a few other places I found in the existing CPython codebase that could benefit from this set of functions:

• the GC young.count allocation counter, which is purely heuristic and is already read with relaxed loads
• lru_cache hit/miss statistics; the counters are already read with relaxed loads
• the dict shared-keys incref which may only run while the caller already keeps the keys alive, and the final decref provides the ordering needed to free them
• the module index allocator and the asyncio task name counter, where only uniqueness of the returned value matters

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

Linked PRs

• gh-155630

首先检查现有的 _Py_atomic_add 函数和关联的 PR gh-155630，然后比较列出的计数器和分配器使用场景与其在不对周围访问进行排序的情况下对原子性的需求。当 relaxed add 变体支持预期的使用场景且不削弱所需的原子行为时，工作即告完成。